CoVEBench is a diagnostic benchmark for compositional instruction-guided video editing created by NJU-LINK. It evaluates video editing models on realistic multi-point instructions that require modifying requested content while preserving unrelated source-video semantics and temporal coherence. The dataset was last updated on June 1, 2026.
Use Cases
- Benchmarking video editing models based on multi-point instructions
- Evaluating temporal coherence preservation in edited videos
- Testing model performance on compositional tasks requiring simultaneous content modification and preservation
- Diagnosing weaknesses in video editing systems based on the three complementary dimensions measured
Strengths
- Focuses on realistic multi-point editing instructions, unlike single-operation benchmarks
- Evaluates three complementary dimensions of video editing performance
- Created by the NJU-LINK research group
Limitations
- Description metadata is limited; actual data quality requires manual inspection after download
- Column-level documentation is absent; field semantics must be inferred after download
- Row count is unknown, which may limit suitability assessment
Provenance
- Source
- NJU-LINK
- Freshness
- Last updated 2026-06-01 16:31:01; freshness should be verified